研究目的
To improve the positioning accuracy and robustness of the monocular ORB-SLAM2 system in dynamic environments and sequences without loops by enhancing feature point selection.
研究成果
The improved system significantly enhances positioning accuracy and robustness in dynamic environments and sequences without loops, with minimal impact on real-time performance. Future work should focus on adaptive parameters and increasing feature point density for better performance.
研究不足
The system uses fixed parameters (e.g., gray scale threshold, cluster numbers) which may not adapt to extreme conditions like strong rotations or changes in light, potentially leading to tracking loss. It cannot completely eliminate scale drift in long-distance sequences without loops.
1：Experimental Design and Method Selection:
The study modifies the ORB-SLAM2 system by incorporating optical flow concepts and K-Means clustering for feature point classification. It uses one-dimensional and two-dimensional K-Means to classify matched feature point pairs based on parameters like θ, driftx, and drifty to select qualified points for camera pose calculation.
2：Sample Selection and Data Sources:
Open-source datasets TUM and KITTI are used for evaluation, including sequences with and without moving objects, rotations, and loops.
3：List of Experimental Equipment and Materials:
A virtual machine running Ubuntu Linux 14.04 on a MacBook Pro (2.4 GHz Intel Core i5, 8GB 1600MHz DDR3) with 4 cores and 4GB memory assigned.
4：04 on a MacBook Pro (4 GHz Intel Core i5, 8GB 1600MHz DDR3) with 4 cores and 4GB memory assigned.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: The system processes frames, extracts and matches feature points, applies K-Means classification to select points, and calculates camera pose. Tests are run multiple times on different sequences to measure accuracy and running time.
5：Data Analysis Methods:
Translation and rotation errors are calculated by comparing system trajectories with ground truth. Statistical analysis includes mean, standard deviation, min, and max values for running time and errors.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
